- From: Martin Thomson <martin.thomson@gmail.com>
- Date: Mon, 8 Jun 2015 10:52:16 -0700
- To: Patrick McManus <mcmanus@ducksong.com>
- Cc: Ryan Hamilton <rch@google.com>, Mark Nottingham <mnot@mnot.net>, HTTP Working Group <ietf-http-wg@w3.org>
I think that this is something that can be addressed by applying CORS, or its model. The point of CORS is to determine that a request target is willing to accept the request (and in this situation, the attacker is the one answering that question). I think that the <1024 solution is hacky as hell, but better than anything I can think of. My alternatives: . Require an HTTP/2 stack to strip Alt-Svc on the way out, unless it is specially authorized. That's predicated on h2 implementers seeing this advice and it might already be too late for that. . Insist on the header field being proposed by resources under /.well-known/. The frame is superior in this sense; hosted resources won't be able to insert those without authorization. The /.well-known/ option might work as a means of breaking out of the 1024-port block. On 8 June 2015 at 10:32, Patrick McManus <mcmanus@ducksong.com> wrote: > yes, with usual caveat that the delay applies to async bg stuff and not qoe > beyond opportunity cost.. so somewhat less impt. > > On Mon, Jun 8, 2015 at 11:59 AM, Ryan Hamilton <rch@google.com> wrote: >> >> FWIW, In the chromium implementation of Alternate-Protocol (and now >> Alt-Svc) we require the new port to be < 1024 if the original port was < >> 1024. >> >> Would CORS incur a latency penalty in these cases? Would we need to wait >> for a CORS request to come back before we could use Alt-Svc? >> >> On Mon, Jun 8, 2015 at 6:11 AM, Patrick McManus <mcmanus@ducksong.com> >> wrote: >>> >>> its not optimal, but I would consider some kind of CORS mechanism (or >>> more likely, CORS :)) here as part of the alt-svc establishment. >>> >>> relatedly I've heard concerns about even cross host with the cert check >>> in environments with broad alternates - and the feeling that this bypasses >>> the spirit of CORS. (though I disagree on that count, I do understand it). >>> >>> >>> >>> On Sun, Jun 7, 2015 at 9:46 PM, Mark Nottingham <mnot@mnot.net> wrote: >>>> >>>> <https://github.com/httpwg/http-extensions/issues/73> >>>> >>>> This issue asks if allowing a header to advertise an alternative on >>>> another port (but still on the same host) is adequate, since in some shared >>>> hosting environments, users will have the ability to add response headers, >>>> as well as listen on other ports. >>>> >>>> Erik has suggested in the issue that it might be helpful to limit these >>>> to privileged ports — i.e., those lower than 1024. I'm assuming such a >>>> restriction would be in place if the origin port were also privileged. >>>> >>>> What do people think? >>>> >>>> -- >>>> Mark Nottingham https://www.mnot.net/ >>>> >>>> >>> >> >
Received on Monday, 8 June 2015 17:52:43 UTC